push: refactor refspec_append_mapped() for subsequent leak-fix
The set_refspecs() caller of refspec_append_mapped() (added in [1]) left open the question[2] of whether the "remote" we lazily fetch might be NULL in the "[...]uniquely name our ref?" case, as remote_get() can return NULL. If we got past the "[...]uniquely name our ref?" case we'd have already segfaulted if we tried to dereference it as "remote->push.nr". In these cases the config mechanism & previous remote validation will have bailed out earlier. Let's refactor this code to clarify that, we'll now BUG() out if we can't get a "remote", and will no longer retrieve it for these common cases where we don't need it. 1. ca02465b413 (push: use remote.$name.push as a refmap, 2013-12-03) 2. https://lore.kernel.org/git/c0c07b89-7eaf-21cd-748e-e14ea57f09fd@web.de/ Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>